Table 6.

The effects of prediction batching.

	P	R	F1	Tokens	Cost
Batches of 50 data points	0.85	0.78	0.80	71,600	$1.43
One per prompt	0.87	0.79	0.81	897,751	$17.96

The table shows performance of the text-davinci-003 model in terms of Precision (P), Recall (R) and micro F₁ measure (F₁). The Tokens column reports the number of tokens each approach exchanged with the model. The Cost column estimates the cost of such exchange given the OpenAI's pricing scheme for the text-davinci-003 model as of the publication of this paper.